create_branch: move msg setup closer to point of use
[git/git-svn.git] / read-cache.c
blobe44775182392c5672fd6ec84eb954db339e5a1ea
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "tempfile.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "refs.h"
12 #include "dir.h"
13 #include "tree.h"
14 #include "commit.h"
15 #include "blob.h"
16 #include "resolve-undo.h"
17 #include "strbuf.h"
18 #include "varint.h"
19 #include "split-index.h"
20 #include "utf8.h"
22 /* Mask for the name length in ce_flags in the on-disk index */
24 #define CE_NAMEMASK (0x0fff)
26 /* Index extensions.
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
39 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
41 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
42 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
46 struct index_state the_index;
47 static const char *alternate_index_output;
49 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
51 istate->cache[nr] = ce;
52 add_name_hash(istate, ce);
55 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
57 struct cache_entry *old = istate->cache[nr];
59 replace_index_entry_in_base(istate, old, ce);
60 remove_name_hash(istate, old);
61 free(old);
62 set_index_entry(istate, nr, ce);
63 ce->ce_flags |= CE_UPDATE_IN_BASE;
64 istate->cache_changed |= CE_ENTRY_CHANGED;
67 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
74 new->ce_flags &= ~CE_HASHED;
75 new->ce_namelen = namelen;
76 new->index = 0;
77 memcpy(new->name, new_name, namelen + 1);
79 cache_tree_invalidate_path(istate, old->name);
80 untracked_cache_remove_from_index(istate, old->name);
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
85 void fill_stat_data(struct stat_data *sd, struct stat *st)
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
98 int match_stat_data(const struct stat_data *sd, struct stat *st)
100 int changed = 0;
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
108 #ifdef USE_NSEC
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
114 #endif
116 if (check_stat) {
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
124 #ifdef USE_STDEV
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
132 #endif
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
137 return changed;
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
145 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
147 fill_stat_data(&ce->ce_stat_data, st);
149 if (assume_unchanged)
150 ce->ce_flags |= CE_VALID;
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
156 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
158 int match = -1;
159 int fd = git_open_cloexec(ce->name, O_RDONLY);
161 if (fd >= 0) {
162 unsigned char sha1[20];
163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
164 match = hashcmp(sha1, ce->oid.hash);
165 /* index_fd() closed the file descriptor already */
167 return match;
170 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
172 int match = -1;
173 void *buffer;
174 unsigned long size;
175 enum object_type type;
176 struct strbuf sb = STRBUF_INIT;
178 if (strbuf_readlink(&sb, ce->name, expected_size))
179 return -1;
181 buffer = read_sha1_file(ce->oid.hash, &type, &size);
182 if (buffer) {
183 if (size == sb.len)
184 match = memcmp(buffer, sb.buf, size);
185 free(buffer);
187 strbuf_release(&sb);
188 return match;
191 static int ce_compare_gitlink(const struct cache_entry *ce)
193 unsigned char sha1[20];
196 * We don't actually require that the .git directory
197 * under GITLINK directory be a valid git directory. It
198 * might even be missing (in case nobody populated that
199 * sub-project).
201 * If so, we consider it always to match.
203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
204 return 0;
205 return hashcmp(sha1, ce->oid.hash);
208 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
210 switch (st->st_mode & S_IFMT) {
211 case S_IFREG:
212 if (ce_compare_data(ce, st))
213 return DATA_CHANGED;
214 break;
215 case S_IFLNK:
216 if (ce_compare_link(ce, xsize_t(st->st_size)))
217 return DATA_CHANGED;
218 break;
219 case S_IFDIR:
220 if (S_ISGITLINK(ce->ce_mode))
221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
222 default:
223 return TYPE_CHANGED;
225 return 0;
228 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
230 unsigned int changed = 0;
232 if (ce->ce_flags & CE_REMOVE)
233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
235 switch (ce->ce_mode & S_IFMT) {
236 case S_IFREG:
237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
238 /* We consider only the owner x bit to be relevant for
239 * "mode changes"
241 if (trust_executable_bit &&
242 (0100 & (ce->ce_mode ^ st->st_mode)))
243 changed |= MODE_CHANGED;
244 break;
245 case S_IFLNK:
246 if (!S_ISLNK(st->st_mode) &&
247 (has_symlinks || !S_ISREG(st->st_mode)))
248 changed |= TYPE_CHANGED;
249 break;
250 case S_IFGITLINK:
251 /* We ignore most of the st_xxx fields for gitlinks */
252 if (!S_ISDIR(st->st_mode))
253 changed |= TYPE_CHANGED;
254 else if (ce_compare_gitlink(ce))
255 changed |= DATA_CHANGED;
256 return changed;
257 default:
258 die("internal error: ce_mode is %o", ce->ce_mode);
261 changed |= match_stat_data(&ce->ce_stat_data, st);
263 /* Racily smudged entry? */
264 if (!ce->ce_stat_data.sd_size) {
265 if (!is_empty_blob_sha1(ce->oid.hash))
266 changed |= DATA_CHANGED;
269 return changed;
272 static int is_racy_stat(const struct index_state *istate,
273 const struct stat_data *sd)
275 return (istate->timestamp.sec &&
276 #ifdef USE_NSEC
277 /* nanosecond timestamped files can also be racy! */
278 (istate->timestamp.sec < sd->sd_mtime.sec ||
279 (istate->timestamp.sec == sd->sd_mtime.sec &&
280 istate->timestamp.nsec <= sd->sd_mtime.nsec))
281 #else
282 istate->timestamp.sec <= sd->sd_mtime.sec
283 #endif
287 static int is_racy_timestamp(const struct index_state *istate,
288 const struct cache_entry *ce)
290 return (!S_ISGITLINK(ce->ce_mode) &&
291 is_racy_stat(istate, &ce->ce_stat_data));
294 int match_stat_data_racy(const struct index_state *istate,
295 const struct stat_data *sd, struct stat *st)
297 if (is_racy_stat(istate, sd))
298 return MTIME_CHANGED;
299 return match_stat_data(sd, st);
302 int ie_match_stat(const struct index_state *istate,
303 const struct cache_entry *ce, struct stat *st,
304 unsigned int options)
306 unsigned int changed;
307 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
312 * If it's marked as always valid in the index, it's
313 * valid whatever the checked-out copy says.
315 * skip-worktree has the same effect with higher precedence
317 if (!ignore_skip_worktree && ce_skip_worktree(ce))
318 return 0;
319 if (!ignore_valid && (ce->ce_flags & CE_VALID))
320 return 0;
323 * Intent-to-add entries have not been added, so the index entry
324 * by definition never matches what is in the work tree until it
325 * actually gets added.
327 if (ce_intent_to_add(ce))
328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
330 changed = ce_match_stat_basic(ce, st);
333 * Within 1 second of this sequence:
334 * echo xyzzy >file && git-update-index --add file
335 * running this command:
336 * echo frotz >file
337 * would give a falsely clean cache entry. The mtime and
338 * length match the cache, and other stat fields do not change.
340 * We could detect this at update-index time (the cache entry
341 * being registered/updated records the same time as "now")
342 * and delay the return from git-update-index, but that would
343 * effectively mean we can make at most one commit per second,
344 * which is not acceptable. Instead, we check cache entries
345 * whose mtime are the same as the index file timestamp more
346 * carefully than others.
348 if (!changed && is_racy_timestamp(istate, ce)) {
349 if (assume_racy_is_modified)
350 changed |= DATA_CHANGED;
351 else
352 changed |= ce_modified_check_fs(ce, st);
355 return changed;
358 int ie_modified(const struct index_state *istate,
359 const struct cache_entry *ce,
360 struct stat *st, unsigned int options)
362 int changed, changed_fs;
364 changed = ie_match_stat(istate, ce, st, options);
365 if (!changed)
366 return 0;
368 * If the mode or type has changed, there's no point in trying
369 * to refresh the entry - it's not going to match
371 if (changed & (MODE_CHANGED | TYPE_CHANGED))
372 return changed;
375 * Immediately after read-tree or update-index --cacheinfo,
376 * the length field is zero, as we have never even read the
377 * lstat(2) information once, and we cannot trust DATA_CHANGED
378 * returned by ie_match_stat() which in turn was returned by
379 * ce_match_stat_basic() to signal that the filesize of the
380 * blob changed. We have to actually go to the filesystem to
381 * see if the contents match, and if so, should answer "unchanged".
383 * The logic does not apply to gitlinks, as ce_match_stat_basic()
384 * already has checked the actual HEAD from the filesystem in the
385 * subproject. If ie_match_stat() already said it is different,
386 * then we know it is.
388 if ((changed & DATA_CHANGED) &&
389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
390 return changed;
392 changed_fs = ce_modified_check_fs(ce, st);
393 if (changed_fs)
394 return changed | changed_fs;
395 return 0;
398 int base_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
401 unsigned char c1, c2;
402 int len = len1 < len2 ? len1 : len2;
403 int cmp;
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 c1 = name1[len];
409 c2 = name2[len];
410 if (!c1 && S_ISDIR(mode1))
411 c1 = '/';
412 if (!c2 && S_ISDIR(mode2))
413 c2 = '/';
414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
418 * df_name_compare() is identical to base_name_compare(), except it
419 * compares conflicting directory/file entries as equal. Note that
420 * while a directory name compares as equal to a regular file, they
421 * then individually compare _differently_ to a filename that has
422 * a dot after the basename (because '\0' < '.' < '/').
424 * This is used by routines that want to traverse the git namespace
425 * but then handle conflicting entries together when possible.
427 int df_name_compare(const char *name1, int len1, int mode1,
428 const char *name2, int len2, int mode2)
430 int len = len1 < len2 ? len1 : len2, cmp;
431 unsigned char c1, c2;
433 cmp = memcmp(name1, name2, len);
434 if (cmp)
435 return cmp;
436 /* Directories and files compare equal (same length, same name) */
437 if (len1 == len2)
438 return 0;
439 c1 = name1[len];
440 if (!c1 && S_ISDIR(mode1))
441 c1 = '/';
442 c2 = name2[len];
443 if (!c2 && S_ISDIR(mode2))
444 c2 = '/';
445 if (c1 == '/' && !c2)
446 return 0;
447 if (c2 == '/' && !c1)
448 return 0;
449 return c1 - c2;
452 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
454 size_t min_len = (len1 < len2) ? len1 : len2;
455 int cmp = memcmp(name1, name2, min_len);
456 if (cmp)
457 return cmp;
458 if (len1 < len2)
459 return -1;
460 if (len1 > len2)
461 return 1;
462 return 0;
465 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
467 int cmp;
469 cmp = name_compare(name1, len1, name2, len2);
470 if (cmp)
471 return cmp;
473 if (stage1 < stage2)
474 return -1;
475 if (stage1 > stage2)
476 return 1;
477 return 0;
480 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
482 int first, last;
484 first = 0;
485 last = istate->cache_nr;
486 while (last > first) {
487 int next = (last + first) >> 1;
488 struct cache_entry *ce = istate->cache[next];
489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
490 if (!cmp)
491 return next;
492 if (cmp < 0) {
493 last = next;
494 continue;
496 first = next+1;
498 return -first-1;
501 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
503 return index_name_stage_pos(istate, name, namelen, 0);
506 int remove_index_entry_at(struct index_state *istate, int pos)
508 struct cache_entry *ce = istate->cache[pos];
510 record_resolve_undo(istate, ce);
511 remove_name_hash(istate, ce);
512 save_or_free_index_entry(istate, ce);
513 istate->cache_changed |= CE_ENTRY_REMOVED;
514 istate->cache_nr--;
515 if (pos >= istate->cache_nr)
516 return 0;
517 memmove(istate->cache + pos,
518 istate->cache + pos + 1,
519 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
520 return 1;
524 * Remove all cache entries marked for removal, that is where
525 * CE_REMOVE is set in ce_flags. This is much more effective than
526 * calling remove_index_entry_at() for each entry to be removed.
528 void remove_marked_cache_entries(struct index_state *istate)
530 struct cache_entry **ce_array = istate->cache;
531 unsigned int i, j;
533 for (i = j = 0; i < istate->cache_nr; i++) {
534 if (ce_array[i]->ce_flags & CE_REMOVE) {
535 remove_name_hash(istate, ce_array[i]);
536 save_or_free_index_entry(istate, ce_array[i]);
538 else
539 ce_array[j++] = ce_array[i];
541 if (j == istate->cache_nr)
542 return;
543 istate->cache_changed |= CE_ENTRY_REMOVED;
544 istate->cache_nr = j;
547 int remove_file_from_index(struct index_state *istate, const char *path)
549 int pos = index_name_pos(istate, path, strlen(path));
550 if (pos < 0)
551 pos = -pos-1;
552 cache_tree_invalidate_path(istate, path);
553 untracked_cache_remove_from_index(istate, path);
554 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
555 remove_index_entry_at(istate, pos);
556 return 0;
559 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
561 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
564 static int index_name_pos_also_unmerged(struct index_state *istate,
565 const char *path, int namelen)
567 int pos = index_name_pos(istate, path, namelen);
568 struct cache_entry *ce;
570 if (pos >= 0)
571 return pos;
573 /* maybe unmerged? */
574 pos = -1 - pos;
575 if (pos >= istate->cache_nr ||
576 compare_name((ce = istate->cache[pos]), path, namelen))
577 return -1;
579 /* order of preference: stage 2, 1, 3 */
580 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
581 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
582 !compare_name(ce, path, namelen))
583 pos++;
584 return pos;
587 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
589 int len = ce_namelen(ce);
590 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
594 * If we add a filename that aliases in the cache, we will use the
595 * name that we already have - but we don't want to update the same
596 * alias twice, because that implies that there were actually two
597 * different files with aliasing names!
599 * So we use the CE_ADDED flag to verify that the alias was an old
600 * one before we accept it as
602 static struct cache_entry *create_alias_ce(struct index_state *istate,
603 struct cache_entry *ce,
604 struct cache_entry *alias)
606 int len;
607 struct cache_entry *new;
609 if (alias->ce_flags & CE_ADDED)
610 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
612 /* Ok, create the new entry using the name of the existing alias */
613 len = ce_namelen(alias);
614 new = xcalloc(1, cache_entry_size(len));
615 memcpy(new->name, alias->name, len);
616 copy_cache_entry(new, ce);
617 save_or_free_index_entry(istate, ce);
618 return new;
621 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
623 unsigned char sha1[20];
624 if (write_sha1_file("", 0, blob_type, sha1))
625 die("cannot create an empty blob in the object database");
626 hashcpy(ce->oid.hash, sha1);
629 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
631 int size, namelen, was_same;
632 mode_t st_mode = st->st_mode;
633 struct cache_entry *ce, *alias;
634 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
635 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
636 int pretend = flags & ADD_CACHE_PRETEND;
637 int intent_only = flags & ADD_CACHE_INTENT;
638 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
639 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
641 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
642 return error("%s: can only add regular files, symbolic links or git-directories", path);
644 namelen = strlen(path);
645 if (S_ISDIR(st_mode)) {
646 while (namelen && path[namelen-1] == '/')
647 namelen--;
649 size = cache_entry_size(namelen);
650 ce = xcalloc(1, size);
651 memcpy(ce->name, path, namelen);
652 ce->ce_namelen = namelen;
653 if (!intent_only)
654 fill_stat_cache_info(ce, st);
655 else
656 ce->ce_flags |= CE_INTENT_TO_ADD;
659 if (trust_executable_bit && has_symlinks) {
660 ce->ce_mode = create_ce_mode(st_mode);
661 } else {
662 /* If there is an existing entry, pick the mode bits and type
663 * from it, otherwise assume unexecutable regular file.
665 struct cache_entry *ent;
666 int pos = index_name_pos_also_unmerged(istate, path, namelen);
668 ent = (0 <= pos) ? istate->cache[pos] : NULL;
669 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
672 /* When core.ignorecase=true, determine if a directory of the same name but differing
673 * case already exists within the Git repository. If it does, ensure the directory
674 * case of the file being added to the repository matches (is folded into) the existing
675 * entry's directory case.
677 if (ignore_case) {
678 adjust_dirname_case(istate, ce->name);
681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
683 /* Nothing changed, really */
684 if (!S_ISGITLINK(alias->ce_mode))
685 ce_mark_uptodate(alias);
686 alias->ce_flags |= CE_ADDED;
688 free(ce);
689 return 0;
691 if (!intent_only) {
692 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
693 free(ce);
694 return error("unable to index file %s", path);
696 } else
697 set_object_name_for_intent_to_add_entry(ce);
699 if (ignore_case && alias && different_name(ce, alias))
700 ce = create_alias_ce(istate, ce, alias);
701 ce->ce_flags |= CE_ADDED;
703 /* It was suspected to be racily clean, but it turns out to be Ok */
704 was_same = (alias &&
705 !ce_stage(alias) &&
706 !oidcmp(&alias->oid, &ce->oid) &&
707 ce->ce_mode == alias->ce_mode);
709 if (pretend)
710 free(ce);
711 else if (add_index_entry(istate, ce, add_option)) {
712 free(ce);
713 return error("unable to add %s to index", path);
715 if (verbose && !was_same)
716 printf("add '%s'\n", path);
717 return 0;
720 int add_file_to_index(struct index_state *istate, const char *path, int flags)
722 struct stat st;
723 if (lstat(path, &st))
724 die_errno("unable to stat '%s'", path);
725 return add_to_index(istate, path, &st, flags);
728 struct cache_entry *make_cache_entry(unsigned int mode,
729 const unsigned char *sha1, const char *path, int stage,
730 unsigned int refresh_options)
732 int size, len;
733 struct cache_entry *ce, *ret;
735 if (!verify_path(path)) {
736 error("Invalid path '%s'", path);
737 return NULL;
740 len = strlen(path);
741 size = cache_entry_size(len);
742 ce = xcalloc(1, size);
744 hashcpy(ce->oid.hash, sha1);
745 memcpy(ce->name, path, len);
746 ce->ce_flags = create_ce_flags(stage);
747 ce->ce_namelen = len;
748 ce->ce_mode = create_ce_mode(mode);
750 ret = refresh_cache_entry(ce, refresh_options);
751 if (ret != ce)
752 free(ce);
753 return ret;
757 * Chmod an index entry with either +x or -x.
759 * Returns -1 if the chmod for the particular cache entry failed (if it's
760 * not a regular file), -2 if an invalid flip argument is passed in, 0
761 * otherwise.
763 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
764 char flip)
766 if (!S_ISREG(ce->ce_mode))
767 return -1;
768 switch (flip) {
769 case '+':
770 ce->ce_mode |= 0111;
771 break;
772 case '-':
773 ce->ce_mode &= ~0111;
774 break;
775 default:
776 return -2;
778 cache_tree_invalidate_path(istate, ce->name);
779 ce->ce_flags |= CE_UPDATE_IN_BASE;
780 istate->cache_changed |= CE_ENTRY_CHANGED;
782 return 0;
785 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
787 int len = ce_namelen(a);
788 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
792 * We fundamentally don't like some paths: we don't want
793 * dot or dot-dot anywhere, and for obvious reasons don't
794 * want to recurse into ".git" either.
796 * Also, we don't want double slashes or slashes at the
797 * end that can make pathnames ambiguous.
799 static int verify_dotfile(const char *rest)
802 * The first character was '.', but that
803 * has already been discarded, we now test
804 * the rest.
807 /* "." is not allowed */
808 if (*rest == '\0' || is_dir_sep(*rest))
809 return 0;
811 switch (*rest) {
813 * ".git" followed by NUL or slash is bad. This
814 * shares the path end test with the ".." case.
816 case 'g':
817 case 'G':
818 if (rest[1] != 'i' && rest[1] != 'I')
819 break;
820 if (rest[2] != 't' && rest[2] != 'T')
821 break;
822 rest += 2;
823 /* fallthrough */
824 case '.':
825 if (rest[1] == '\0' || is_dir_sep(rest[1]))
826 return 0;
828 return 1;
831 int verify_path(const char *path)
833 char c;
835 if (has_dos_drive_prefix(path))
836 return 0;
838 goto inside;
839 for (;;) {
840 if (!c)
841 return 1;
842 if (is_dir_sep(c)) {
843 inside:
844 if (protect_hfs && is_hfs_dotgit(path))
845 return 0;
846 if (protect_ntfs && is_ntfs_dotgit(path))
847 return 0;
848 c = *path++;
849 if ((c == '.' && !verify_dotfile(path)) ||
850 is_dir_sep(c) || c == '\0')
851 return 0;
853 c = *path++;
858 * Do we have another file that has the beginning components being a
859 * proper superset of the name we're trying to add?
861 static int has_file_name(struct index_state *istate,
862 const struct cache_entry *ce, int pos, int ok_to_replace)
864 int retval = 0;
865 int len = ce_namelen(ce);
866 int stage = ce_stage(ce);
867 const char *name = ce->name;
869 while (pos < istate->cache_nr) {
870 struct cache_entry *p = istate->cache[pos++];
872 if (len >= ce_namelen(p))
873 break;
874 if (memcmp(name, p->name, len))
875 break;
876 if (ce_stage(p) != stage)
877 continue;
878 if (p->name[len] != '/')
879 continue;
880 if (p->ce_flags & CE_REMOVE)
881 continue;
882 retval = -1;
883 if (!ok_to_replace)
884 break;
885 remove_index_entry_at(istate, --pos);
887 return retval;
891 * Do we have another file with a pathname that is a proper
892 * subset of the name we're trying to add?
894 static int has_dir_name(struct index_state *istate,
895 const struct cache_entry *ce, int pos, int ok_to_replace)
897 int retval = 0;
898 int stage = ce_stage(ce);
899 const char *name = ce->name;
900 const char *slash = name + ce_namelen(ce);
902 for (;;) {
903 int len;
905 for (;;) {
906 if (*--slash == '/')
907 break;
908 if (slash <= ce->name)
909 return retval;
911 len = slash - name;
913 pos = index_name_stage_pos(istate, name, len, stage);
914 if (pos >= 0) {
916 * Found one, but not so fast. This could
917 * be a marker that says "I was here, but
918 * I am being removed". Such an entry is
919 * not a part of the resulting tree, and
920 * it is Ok to have a directory at the same
921 * path.
923 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
924 retval = -1;
925 if (!ok_to_replace)
926 break;
927 remove_index_entry_at(istate, pos);
928 continue;
931 else
932 pos = -pos-1;
935 * Trivial optimization: if we find an entry that
936 * already matches the sub-directory, then we know
937 * we're ok, and we can exit.
939 while (pos < istate->cache_nr) {
940 struct cache_entry *p = istate->cache[pos];
941 if ((ce_namelen(p) <= len) ||
942 (p->name[len] != '/') ||
943 memcmp(p->name, name, len))
944 break; /* not our subdirectory */
945 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
947 * p is at the same stage as our entry, and
948 * is a subdirectory of what we are looking
949 * at, so we cannot have conflicts at our
950 * level or anything shorter.
952 return retval;
953 pos++;
956 return retval;
959 /* We may be in a situation where we already have path/file and path
960 * is being added, or we already have path and path/file is being
961 * added. Either one would result in a nonsense tree that has path
962 * twice when git-write-tree tries to write it out. Prevent it.
964 * If ok-to-replace is specified, we remove the conflicting entries
965 * from the cache so the caller should recompute the insert position.
966 * When this happens, we return non-zero.
968 static int check_file_directory_conflict(struct index_state *istate,
969 const struct cache_entry *ce,
970 int pos, int ok_to_replace)
972 int retval;
975 * When ce is an "I am going away" entry, we allow it to be added
977 if (ce->ce_flags & CE_REMOVE)
978 return 0;
981 * We check if the path is a sub-path of a subsequent pathname
982 * first, since removing those will not change the position
983 * in the array.
985 retval = has_file_name(istate, ce, pos, ok_to_replace);
988 * Then check if the path might have a clashing sub-directory
989 * before it.
991 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
994 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
996 int pos;
997 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
998 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
999 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1000 int new_only = option & ADD_CACHE_NEW_ONLY;
1002 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1003 cache_tree_invalidate_path(istate, ce->name);
1004 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1006 /* existing match? Just replace it. */
1007 if (pos >= 0) {
1008 if (!new_only)
1009 replace_index_entry(istate, pos, ce);
1010 return 0;
1012 pos = -pos-1;
1014 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1015 untracked_cache_add_to_index(istate, ce->name);
1018 * Inserting a merged entry ("stage 0") into the index
1019 * will always replace all non-merged entries..
1021 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1022 while (ce_same_name(istate->cache[pos], ce)) {
1023 ok_to_add = 1;
1024 if (!remove_index_entry_at(istate, pos))
1025 break;
1029 if (!ok_to_add)
1030 return -1;
1031 if (!verify_path(ce->name))
1032 return error("Invalid path '%s'", ce->name);
1034 if (!skip_df_check &&
1035 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1036 if (!ok_to_replace)
1037 return error("'%s' appears as both a file and as a directory",
1038 ce->name);
1039 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1040 pos = -pos-1;
1042 return pos + 1;
1045 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1047 int pos;
1049 if (option & ADD_CACHE_JUST_APPEND)
1050 pos = istate->cache_nr;
1051 else {
1052 int ret;
1053 ret = add_index_entry_with_check(istate, ce, option);
1054 if (ret <= 0)
1055 return ret;
1056 pos = ret - 1;
1059 /* Make sure the array is big enough .. */
1060 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1062 /* Add it in.. */
1063 istate->cache_nr++;
1064 if (istate->cache_nr > pos + 1)
1065 memmove(istate->cache + pos + 1,
1066 istate->cache + pos,
1067 (istate->cache_nr - pos - 1) * sizeof(ce));
1068 set_index_entry(istate, pos, ce);
1069 istate->cache_changed |= CE_ENTRY_ADDED;
1070 return 0;
1074 * "refresh" does not calculate a new sha1 file or bring the
1075 * cache up-to-date for mode/content changes. But what it
1076 * _does_ do is to "re-match" the stat information of a file
1077 * with the cache, so that you can refresh the cache for a
1078 * file that hasn't been changed but where the stat entry is
1079 * out of date.
1081 * For example, you'd want to do this after doing a "git-read-tree",
1082 * to link up the stat cache details with the proper files.
1084 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1085 struct cache_entry *ce,
1086 unsigned int options, int *err,
1087 int *changed_ret)
1089 struct stat st;
1090 struct cache_entry *updated;
1091 int changed, size;
1092 int refresh = options & CE_MATCH_REFRESH;
1093 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1094 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1095 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1097 if (!refresh || ce_uptodate(ce))
1098 return ce;
1101 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1102 * that the change to the work tree does not matter and told
1103 * us not to worry.
1105 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1106 ce_mark_uptodate(ce);
1107 return ce;
1109 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1110 ce_mark_uptodate(ce);
1111 return ce;
1114 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1115 if (ignore_missing)
1116 return ce;
1117 if (err)
1118 *err = ENOENT;
1119 return NULL;
1122 if (lstat(ce->name, &st) < 0) {
1123 if (ignore_missing && errno == ENOENT)
1124 return ce;
1125 if (err)
1126 *err = errno;
1127 return NULL;
1130 changed = ie_match_stat(istate, ce, &st, options);
1131 if (changed_ret)
1132 *changed_ret = changed;
1133 if (!changed) {
1135 * The path is unchanged. If we were told to ignore
1136 * valid bit, then we did the actual stat check and
1137 * found that the entry is unmodified. If the entry
1138 * is not marked VALID, this is the place to mark it
1139 * valid again, under "assume unchanged" mode.
1141 if (ignore_valid && assume_unchanged &&
1142 !(ce->ce_flags & CE_VALID))
1143 ; /* mark this one VALID again */
1144 else {
1146 * We do not mark the index itself "modified"
1147 * because CE_UPTODATE flag is in-core only;
1148 * we are not going to write this change out.
1150 if (!S_ISGITLINK(ce->ce_mode))
1151 ce_mark_uptodate(ce);
1152 return ce;
1156 if (ie_modified(istate, ce, &st, options)) {
1157 if (err)
1158 *err = EINVAL;
1159 return NULL;
1162 size = ce_size(ce);
1163 updated = xmalloc(size);
1164 memcpy(updated, ce, size);
1165 fill_stat_cache_info(updated, &st);
1167 * If ignore_valid is not set, we should leave CE_VALID bit
1168 * alone. Otherwise, paths marked with --no-assume-unchanged
1169 * (i.e. things to be edited) will reacquire CE_VALID bit
1170 * automatically, which is not really what we want.
1172 if (!ignore_valid && assume_unchanged &&
1173 !(ce->ce_flags & CE_VALID))
1174 updated->ce_flags &= ~CE_VALID;
1176 /* istate->cache_changed is updated in the caller */
1177 return updated;
1180 static void show_file(const char * fmt, const char * name, int in_porcelain,
1181 int * first, const char *header_msg)
1183 if (in_porcelain && *first && header_msg) {
1184 printf("%s\n", header_msg);
1185 *first = 0;
1187 printf(fmt, name);
1190 int refresh_index(struct index_state *istate, unsigned int flags,
1191 const struct pathspec *pathspec,
1192 char *seen, const char *header_msg)
1194 int i;
1195 int has_errors = 0;
1196 int really = (flags & REFRESH_REALLY) != 0;
1197 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1198 int quiet = (flags & REFRESH_QUIET) != 0;
1199 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1200 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1201 int first = 1;
1202 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1203 unsigned int options = (CE_MATCH_REFRESH |
1204 (really ? CE_MATCH_IGNORE_VALID : 0) |
1205 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1206 const char *modified_fmt;
1207 const char *deleted_fmt;
1208 const char *typechange_fmt;
1209 const char *added_fmt;
1210 const char *unmerged_fmt;
1212 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1213 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1214 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1215 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1216 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1217 for (i = 0; i < istate->cache_nr; i++) {
1218 struct cache_entry *ce, *new;
1219 int cache_errno = 0;
1220 int changed = 0;
1221 int filtered = 0;
1223 ce = istate->cache[i];
1224 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1225 continue;
1227 if (pathspec && !ce_path_match(ce, pathspec, seen))
1228 filtered = 1;
1230 if (ce_stage(ce)) {
1231 while ((i < istate->cache_nr) &&
1232 ! strcmp(istate->cache[i]->name, ce->name))
1233 i++;
1234 i--;
1235 if (allow_unmerged)
1236 continue;
1237 if (!filtered)
1238 show_file(unmerged_fmt, ce->name, in_porcelain,
1239 &first, header_msg);
1240 has_errors = 1;
1241 continue;
1244 if (filtered)
1245 continue;
1247 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1248 if (new == ce)
1249 continue;
1250 if (!new) {
1251 const char *fmt;
1253 if (really && cache_errno == EINVAL) {
1254 /* If we are doing --really-refresh that
1255 * means the index is not valid anymore.
1257 ce->ce_flags &= ~CE_VALID;
1258 ce->ce_flags |= CE_UPDATE_IN_BASE;
1259 istate->cache_changed |= CE_ENTRY_CHANGED;
1261 if (quiet)
1262 continue;
1264 if (cache_errno == ENOENT)
1265 fmt = deleted_fmt;
1266 else if (ce_intent_to_add(ce))
1267 fmt = added_fmt; /* must be before other checks */
1268 else if (changed & TYPE_CHANGED)
1269 fmt = typechange_fmt;
1270 else
1271 fmt = modified_fmt;
1272 show_file(fmt,
1273 ce->name, in_porcelain, &first, header_msg);
1274 has_errors = 1;
1275 continue;
1278 replace_index_entry(istate, i, new);
1280 return has_errors;
1283 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1284 unsigned int options)
1286 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1290 /*****************************************************************
1291 * Index File I/O
1292 *****************************************************************/
1294 #define INDEX_FORMAT_DEFAULT 3
1296 static unsigned int get_index_format_default(void)
1298 char *envversion = getenv("GIT_INDEX_VERSION");
1299 char *endp;
1300 int value;
1301 unsigned int version = INDEX_FORMAT_DEFAULT;
1303 if (!envversion) {
1304 if (!git_config_get_int("index.version", &value))
1305 version = value;
1306 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1307 warning(_("index.version set, but the value is invalid.\n"
1308 "Using version %i"), INDEX_FORMAT_DEFAULT);
1309 return INDEX_FORMAT_DEFAULT;
1311 return version;
1314 version = strtoul(envversion, &endp, 10);
1315 if (*endp ||
1316 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1317 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1318 "Using version %i"), INDEX_FORMAT_DEFAULT);
1319 version = INDEX_FORMAT_DEFAULT;
1321 return version;
1325 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1326 * Again - this is just a (very strong in practice) heuristic that
1327 * the inode hasn't changed.
1329 * We save the fields in big-endian order to allow using the
1330 * index file over NFS transparently.
1332 struct ondisk_cache_entry {
1333 struct cache_time ctime;
1334 struct cache_time mtime;
1335 uint32_t dev;
1336 uint32_t ino;
1337 uint32_t mode;
1338 uint32_t uid;
1339 uint32_t gid;
1340 uint32_t size;
1341 unsigned char sha1[20];
1342 uint16_t flags;
1343 char name[FLEX_ARRAY]; /* more */
1347 * This struct is used when CE_EXTENDED bit is 1
1348 * The struct must match ondisk_cache_entry exactly from
1349 * ctime till flags
1351 struct ondisk_cache_entry_extended {
1352 struct cache_time ctime;
1353 struct cache_time mtime;
1354 uint32_t dev;
1355 uint32_t ino;
1356 uint32_t mode;
1357 uint32_t uid;
1358 uint32_t gid;
1359 uint32_t size;
1360 unsigned char sha1[20];
1361 uint16_t flags;
1362 uint16_t flags2;
1363 char name[FLEX_ARRAY]; /* more */
1366 /* These are only used for v3 or lower */
1367 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1368 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1369 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1370 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1371 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1372 ondisk_cache_entry_size(ce_namelen(ce)))
1374 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1376 git_SHA_CTX c;
1377 unsigned char sha1[20];
1378 int hdr_version;
1380 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1381 return error("bad signature");
1382 hdr_version = ntohl(hdr->hdr_version);
1383 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1384 return error("bad index version %d", hdr_version);
1385 git_SHA1_Init(&c);
1386 git_SHA1_Update(&c, hdr, size - 20);
1387 git_SHA1_Final(sha1, &c);
1388 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1389 return error("bad index file sha1 signature");
1390 return 0;
1393 static int read_index_extension(struct index_state *istate,
1394 const char *ext, void *data, unsigned long sz)
1396 switch (CACHE_EXT(ext)) {
1397 case CACHE_EXT_TREE:
1398 istate->cache_tree = cache_tree_read(data, sz);
1399 break;
1400 case CACHE_EXT_RESOLVE_UNDO:
1401 istate->resolve_undo = resolve_undo_read(data, sz);
1402 break;
1403 case CACHE_EXT_LINK:
1404 if (read_link_extension(istate, data, sz))
1405 return -1;
1406 break;
1407 case CACHE_EXT_UNTRACKED:
1408 istate->untracked = read_untracked_extension(data, sz);
1409 break;
1410 default:
1411 if (*ext < 'A' || 'Z' < *ext)
1412 return error("index uses %.4s extension, which we do not understand",
1413 ext);
1414 fprintf(stderr, "ignoring %.4s extension\n", ext);
1415 break;
1417 return 0;
1420 int hold_locked_index(struct lock_file *lk, int lock_flags)
1422 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1425 int read_index(struct index_state *istate)
1427 return read_index_from(istate, get_index_file());
1430 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1431 unsigned int flags,
1432 const char *name,
1433 size_t len)
1435 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1437 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1438 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1439 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1440 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1441 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1442 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1443 ce->ce_mode = get_be32(&ondisk->mode);
1444 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1445 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1446 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1447 ce->ce_flags = flags & ~CE_NAMEMASK;
1448 ce->ce_namelen = len;
1449 ce->index = 0;
1450 hashcpy(ce->oid.hash, ondisk->sha1);
1451 memcpy(ce->name, name, len);
1452 ce->name[len] = '\0';
1453 return ce;
1457 * Adjacent cache entries tend to share the leading paths, so it makes
1458 * sense to only store the differences in later entries. In the v4
1459 * on-disk format of the index, each on-disk cache entry stores the
1460 * number of bytes to be stripped from the end of the previous name,
1461 * and the bytes to append to the result, to come up with its name.
1463 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1465 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1466 size_t len = decode_varint(&cp);
1468 if (name->len < len)
1469 die("malformed name field in the index");
1470 strbuf_remove(name, name->len - len, len);
1471 for (ep = cp; *ep; ep++)
1472 ; /* find the end */
1473 strbuf_add(name, cp, ep - cp);
1474 return (const char *)ep + 1 - cp_;
1477 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1478 unsigned long *ent_size,
1479 struct strbuf *previous_name)
1481 struct cache_entry *ce;
1482 size_t len;
1483 const char *name;
1484 unsigned int flags;
1486 /* On-disk flags are just 16 bits */
1487 flags = get_be16(&ondisk->flags);
1488 len = flags & CE_NAMEMASK;
1490 if (flags & CE_EXTENDED) {
1491 struct ondisk_cache_entry_extended *ondisk2;
1492 int extended_flags;
1493 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1494 extended_flags = get_be16(&ondisk2->flags2) << 16;
1495 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1496 if (extended_flags & ~CE_EXTENDED_FLAGS)
1497 die("Unknown index entry format %08x", extended_flags);
1498 flags |= extended_flags;
1499 name = ondisk2->name;
1501 else
1502 name = ondisk->name;
1504 if (!previous_name) {
1505 /* v3 and earlier */
1506 if (len == CE_NAMEMASK)
1507 len = strlen(name);
1508 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1510 *ent_size = ondisk_ce_size(ce);
1511 } else {
1512 unsigned long consumed;
1513 consumed = expand_name_field(previous_name, name);
1514 ce = cache_entry_from_ondisk(ondisk, flags,
1515 previous_name->buf,
1516 previous_name->len);
1518 *ent_size = (name - ((char *)ondisk)) + consumed;
1520 return ce;
1523 static void check_ce_order(struct index_state *istate)
1525 unsigned int i;
1527 for (i = 1; i < istate->cache_nr; i++) {
1528 struct cache_entry *ce = istate->cache[i - 1];
1529 struct cache_entry *next_ce = istate->cache[i];
1530 int name_compare = strcmp(ce->name, next_ce->name);
1532 if (0 < name_compare)
1533 die("unordered stage entries in index");
1534 if (!name_compare) {
1535 if (!ce_stage(ce))
1536 die("multiple stage entries for merged file '%s'",
1537 ce->name);
1538 if (ce_stage(ce) > ce_stage(next_ce))
1539 die("unordered stage entries for '%s'",
1540 ce->name);
1545 static void tweak_untracked_cache(struct index_state *istate)
1547 switch (git_config_get_untracked_cache()) {
1548 case -1: /* keep: do nothing */
1549 break;
1550 case 0: /* false */
1551 remove_untracked_cache(istate);
1552 break;
1553 case 1: /* true */
1554 add_untracked_cache(istate);
1555 break;
1556 default: /* unknown value: do nothing */
1557 break;
1561 static void tweak_split_index(struct index_state *istate)
1563 switch (git_config_get_split_index()) {
1564 case -1: /* unset: do nothing */
1565 break;
1566 case 0: /* false */
1567 remove_split_index(istate);
1568 break;
1569 case 1: /* true */
1570 add_split_index(istate);
1571 break;
1572 default: /* unknown value: do nothing */
1573 break;
1577 static void post_read_index_from(struct index_state *istate)
1579 check_ce_order(istate);
1580 tweak_untracked_cache(istate);
1581 tweak_split_index(istate);
1584 /* remember to discard_cache() before reading a different cache! */
1585 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1587 int fd, i;
1588 struct stat st;
1589 unsigned long src_offset;
1590 struct cache_header *hdr;
1591 void *mmap;
1592 size_t mmap_size;
1593 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1595 if (istate->initialized)
1596 return istate->cache_nr;
1598 istate->timestamp.sec = 0;
1599 istate->timestamp.nsec = 0;
1600 fd = open(path, O_RDONLY);
1601 if (fd < 0) {
1602 if (!must_exist && errno == ENOENT)
1603 return 0;
1604 die_errno("%s: index file open failed", path);
1607 if (fstat(fd, &st))
1608 die_errno("cannot stat the open index");
1610 mmap_size = xsize_t(st.st_size);
1611 if (mmap_size < sizeof(struct cache_header) + 20)
1612 die("index file smaller than expected");
1614 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1615 if (mmap == MAP_FAILED)
1616 die_errno("unable to map index file");
1617 close(fd);
1619 hdr = mmap;
1620 if (verify_hdr(hdr, mmap_size) < 0)
1621 goto unmap;
1623 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1624 istate->version = ntohl(hdr->hdr_version);
1625 istate->cache_nr = ntohl(hdr->hdr_entries);
1626 istate->cache_alloc = alloc_nr(istate->cache_nr);
1627 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1628 istate->initialized = 1;
1630 if (istate->version == 4)
1631 previous_name = &previous_name_buf;
1632 else
1633 previous_name = NULL;
1635 src_offset = sizeof(*hdr);
1636 for (i = 0; i < istate->cache_nr; i++) {
1637 struct ondisk_cache_entry *disk_ce;
1638 struct cache_entry *ce;
1639 unsigned long consumed;
1641 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1642 ce = create_from_disk(disk_ce, &consumed, previous_name);
1643 set_index_entry(istate, i, ce);
1645 src_offset += consumed;
1647 strbuf_release(&previous_name_buf);
1648 istate->timestamp.sec = st.st_mtime;
1649 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1651 while (src_offset <= mmap_size - 20 - 8) {
1652 /* After an array of active_nr index entries,
1653 * there can be arbitrary number of extended
1654 * sections, each of which is prefixed with
1655 * extension name (4-byte) and section length
1656 * in 4-byte network byte order.
1658 uint32_t extsize;
1659 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1660 extsize = ntohl(extsize);
1661 if (read_index_extension(istate,
1662 (const char *) mmap + src_offset,
1663 (char *) mmap + src_offset + 8,
1664 extsize) < 0)
1665 goto unmap;
1666 src_offset += 8;
1667 src_offset += extsize;
1669 munmap(mmap, mmap_size);
1670 return istate->cache_nr;
1672 unmap:
1673 munmap(mmap, mmap_size);
1674 die("index file corrupt");
1678 * Signal that the shared index is used by updating its mtime.
1680 * This way, shared index can be removed if they have not been used
1681 * for some time.
1683 static void freshen_shared_index(char *base_sha1_hex, int warn)
1685 const char *shared_index = git_path("sharedindex.%s", base_sha1_hex);
1686 if (!check_and_freshen_file(shared_index, 1) && warn)
1687 warning("could not freshen shared index '%s'", shared_index);
1690 int read_index_from(struct index_state *istate, const char *path)
1692 struct split_index *split_index;
1693 int ret;
1694 char *base_sha1_hex;
1695 const char *base_path;
1697 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1698 if (istate->initialized)
1699 return istate->cache_nr;
1701 ret = do_read_index(istate, path, 0);
1703 split_index = istate->split_index;
1704 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1705 post_read_index_from(istate);
1706 return ret;
1709 if (split_index->base)
1710 discard_index(split_index->base);
1711 else
1712 split_index->base = xcalloc(1, sizeof(*split_index->base));
1714 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1715 base_path = git_path("sharedindex.%s", base_sha1_hex);
1716 ret = do_read_index(split_index->base, base_path, 1);
1717 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1718 die("broken index, expect %s in %s, got %s",
1719 base_sha1_hex, base_path,
1720 sha1_to_hex(split_index->base->sha1));
1722 freshen_shared_index(base_sha1_hex, 0);
1723 merge_base_index(istate);
1724 post_read_index_from(istate);
1725 return ret;
1728 int is_index_unborn(struct index_state *istate)
1730 return (!istate->cache_nr && !istate->timestamp.sec);
1733 int discard_index(struct index_state *istate)
1735 int i;
1737 for (i = 0; i < istate->cache_nr; i++) {
1738 if (istate->cache[i]->index &&
1739 istate->split_index &&
1740 istate->split_index->base &&
1741 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1742 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1743 continue;
1744 free(istate->cache[i]);
1746 resolve_undo_clear_index(istate);
1747 istate->cache_nr = 0;
1748 istate->cache_changed = 0;
1749 istate->timestamp.sec = 0;
1750 istate->timestamp.nsec = 0;
1751 free_name_hash(istate);
1752 cache_tree_free(&(istate->cache_tree));
1753 istate->initialized = 0;
1754 free(istate->cache);
1755 istate->cache = NULL;
1756 istate->cache_alloc = 0;
1757 discard_split_index(istate);
1758 free_untracked_cache(istate->untracked);
1759 istate->untracked = NULL;
1760 return 0;
1763 int unmerged_index(const struct index_state *istate)
1765 int i;
1766 for (i = 0; i < istate->cache_nr; i++) {
1767 if (ce_stage(istate->cache[i]))
1768 return 1;
1770 return 0;
1773 #define WRITE_BUFFER_SIZE 8192
1774 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1775 static unsigned long write_buffer_len;
1777 static int ce_write_flush(git_SHA_CTX *context, int fd)
1779 unsigned int buffered = write_buffer_len;
1780 if (buffered) {
1781 git_SHA1_Update(context, write_buffer, buffered);
1782 if (write_in_full(fd, write_buffer, buffered) != buffered)
1783 return -1;
1784 write_buffer_len = 0;
1786 return 0;
1789 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1791 while (len) {
1792 unsigned int buffered = write_buffer_len;
1793 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1794 if (partial > len)
1795 partial = len;
1796 memcpy(write_buffer + buffered, data, partial);
1797 buffered += partial;
1798 if (buffered == WRITE_BUFFER_SIZE) {
1799 write_buffer_len = buffered;
1800 if (ce_write_flush(context, fd))
1801 return -1;
1802 buffered = 0;
1804 write_buffer_len = buffered;
1805 len -= partial;
1806 data = (char *) data + partial;
1808 return 0;
1811 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1812 unsigned int ext, unsigned int sz)
1814 ext = htonl(ext);
1815 sz = htonl(sz);
1816 return ((ce_write(context, fd, &ext, 4) < 0) ||
1817 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1820 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1822 unsigned int left = write_buffer_len;
1824 if (left) {
1825 write_buffer_len = 0;
1826 git_SHA1_Update(context, write_buffer, left);
1829 /* Flush first if not enough space for SHA1 signature */
1830 if (left + 20 > WRITE_BUFFER_SIZE) {
1831 if (write_in_full(fd, write_buffer, left) != left)
1832 return -1;
1833 left = 0;
1836 /* Append the SHA1 signature at the end */
1837 git_SHA1_Final(write_buffer + left, context);
1838 hashcpy(sha1, write_buffer + left);
1839 left += 20;
1840 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1843 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1846 * The only thing we care about in this function is to smudge the
1847 * falsely clean entry due to touch-update-touch race, so we leave
1848 * everything else as they are. We are called for entries whose
1849 * ce_stat_data.sd_mtime match the index file mtime.
1851 * Note that this actually does not do much for gitlinks, for
1852 * which ce_match_stat_basic() always goes to the actual
1853 * contents. The caller checks with is_racy_timestamp() which
1854 * always says "no" for gitlinks, so we are not called for them ;-)
1856 struct stat st;
1858 if (lstat(ce->name, &st) < 0)
1859 return;
1860 if (ce_match_stat_basic(ce, &st))
1861 return;
1862 if (ce_modified_check_fs(ce, &st)) {
1863 /* This is "racily clean"; smudge it. Note that this
1864 * is a tricky code. At first glance, it may appear
1865 * that it can break with this sequence:
1867 * $ echo xyzzy >frotz
1868 * $ git-update-index --add frotz
1869 * $ : >frotz
1870 * $ sleep 3
1871 * $ echo filfre >nitfol
1872 * $ git-update-index --add nitfol
1874 * but it does not. When the second update-index runs,
1875 * it notices that the entry "frotz" has the same timestamp
1876 * as index, and if we were to smudge it by resetting its
1877 * size to zero here, then the object name recorded
1878 * in index is the 6-byte file but the cached stat information
1879 * becomes zero --- which would then match what we would
1880 * obtain from the filesystem next time we stat("frotz").
1882 * However, the second update-index, before calling
1883 * this function, notices that the cached size is 6
1884 * bytes and what is on the filesystem is an empty
1885 * file, and never calls us, so the cached size information
1886 * for "frotz" stays 6 which does not match the filesystem.
1888 ce->ce_stat_data.sd_size = 0;
1892 /* Copy miscellaneous fields but not the name */
1893 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1894 struct cache_entry *ce)
1896 short flags;
1898 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1899 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1900 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1901 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1902 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1903 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1904 ondisk->mode = htonl(ce->ce_mode);
1905 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1906 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1907 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1908 hashcpy(ondisk->sha1, ce->oid.hash);
1910 flags = ce->ce_flags & ~CE_NAMEMASK;
1911 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1912 ondisk->flags = htons(flags);
1913 if (ce->ce_flags & CE_EXTENDED) {
1914 struct ondisk_cache_entry_extended *ondisk2;
1915 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1916 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1917 return ondisk2->name;
1919 else {
1920 return ondisk->name;
1924 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1925 struct strbuf *previous_name)
1927 int size;
1928 struct ondisk_cache_entry *ondisk;
1929 int saved_namelen = saved_namelen; /* compiler workaround */
1930 char *name;
1931 int result;
1933 if (ce->ce_flags & CE_STRIP_NAME) {
1934 saved_namelen = ce_namelen(ce);
1935 ce->ce_namelen = 0;
1938 if (!previous_name) {
1939 size = ondisk_ce_size(ce);
1940 ondisk = xcalloc(1, size);
1941 name = copy_cache_entry_to_ondisk(ondisk, ce);
1942 memcpy(name, ce->name, ce_namelen(ce));
1943 } else {
1944 int common, to_remove, prefix_size;
1945 unsigned char to_remove_vi[16];
1946 for (common = 0;
1947 (ce->name[common] &&
1948 common < previous_name->len &&
1949 ce->name[common] == previous_name->buf[common]);
1950 common++)
1951 ; /* still matching */
1952 to_remove = previous_name->len - common;
1953 prefix_size = encode_varint(to_remove, to_remove_vi);
1955 if (ce->ce_flags & CE_EXTENDED)
1956 size = offsetof(struct ondisk_cache_entry_extended, name);
1957 else
1958 size = offsetof(struct ondisk_cache_entry, name);
1959 size += prefix_size + (ce_namelen(ce) - common + 1);
1961 ondisk = xcalloc(1, size);
1962 name = copy_cache_entry_to_ondisk(ondisk, ce);
1963 memcpy(name, to_remove_vi, prefix_size);
1964 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1966 strbuf_splice(previous_name, common, to_remove,
1967 ce->name + common, ce_namelen(ce) - common);
1969 if (ce->ce_flags & CE_STRIP_NAME) {
1970 ce->ce_namelen = saved_namelen;
1971 ce->ce_flags &= ~CE_STRIP_NAME;
1974 result = ce_write(c, fd, ondisk, size);
1975 free(ondisk);
1976 return result;
1980 * This function verifies if index_state has the correct sha1 of the
1981 * index file. Don't die if we have any other failure, just return 0.
1983 static int verify_index_from(const struct index_state *istate, const char *path)
1985 int fd;
1986 ssize_t n;
1987 struct stat st;
1988 unsigned char sha1[20];
1990 if (!istate->initialized)
1991 return 0;
1993 fd = open(path, O_RDONLY);
1994 if (fd < 0)
1995 return 0;
1997 if (fstat(fd, &st))
1998 goto out;
2000 if (st.st_size < sizeof(struct cache_header) + 20)
2001 goto out;
2003 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2004 if (n != 20)
2005 goto out;
2007 if (hashcmp(istate->sha1, sha1))
2008 goto out;
2010 close(fd);
2011 return 1;
2013 out:
2014 close(fd);
2015 return 0;
2018 static int verify_index(const struct index_state *istate)
2020 return verify_index_from(istate, get_index_file());
2023 static int has_racy_timestamp(struct index_state *istate)
2025 int entries = istate->cache_nr;
2026 int i;
2028 for (i = 0; i < entries; i++) {
2029 struct cache_entry *ce = istate->cache[i];
2030 if (is_racy_timestamp(istate, ce))
2031 return 1;
2033 return 0;
2037 * Opportunistically update the index but do not complain if we can't
2039 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2041 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2042 verify_index(istate) &&
2043 write_locked_index(istate, lockfile, COMMIT_LOCK))
2044 rollback_lock_file(lockfile);
2047 static int do_write_index(struct index_state *istate, int newfd,
2048 int strip_extensions)
2050 git_SHA_CTX c;
2051 struct cache_header hdr;
2052 int i, err, removed, extended, hdr_version;
2053 struct cache_entry **cache = istate->cache;
2054 int entries = istate->cache_nr;
2055 struct stat st;
2056 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2058 for (i = removed = extended = 0; i < entries; i++) {
2059 if (cache[i]->ce_flags & CE_REMOVE)
2060 removed++;
2062 /* reduce extended entries if possible */
2063 cache[i]->ce_flags &= ~CE_EXTENDED;
2064 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2065 extended++;
2066 cache[i]->ce_flags |= CE_EXTENDED;
2070 if (!istate->version) {
2071 istate->version = get_index_format_default();
2072 if (getenv("GIT_TEST_SPLIT_INDEX"))
2073 init_split_index(istate);
2076 /* demote version 3 to version 2 when the latter suffices */
2077 if (istate->version == 3 || istate->version == 2)
2078 istate->version = extended ? 3 : 2;
2080 hdr_version = istate->version;
2082 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2083 hdr.hdr_version = htonl(hdr_version);
2084 hdr.hdr_entries = htonl(entries - removed);
2086 git_SHA1_Init(&c);
2087 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2088 return -1;
2090 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2091 for (i = 0; i < entries; i++) {
2092 struct cache_entry *ce = cache[i];
2093 if (ce->ce_flags & CE_REMOVE)
2094 continue;
2095 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2096 ce_smudge_racily_clean_entry(ce);
2097 if (is_null_oid(&ce->oid)) {
2098 static const char msg[] = "cache entry has null sha1: %s";
2099 static int allow = -1;
2101 if (allow < 0)
2102 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2103 if (allow)
2104 warning(msg, ce->name);
2105 else
2106 return error(msg, ce->name);
2108 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2109 return -1;
2111 strbuf_release(&previous_name_buf);
2113 /* Write extension data here */
2114 if (!strip_extensions && istate->split_index) {
2115 struct strbuf sb = STRBUF_INIT;
2117 err = write_link_extension(&sb, istate) < 0 ||
2118 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2119 sb.len) < 0 ||
2120 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2121 strbuf_release(&sb);
2122 if (err)
2123 return -1;
2125 if (!strip_extensions && istate->cache_tree) {
2126 struct strbuf sb = STRBUF_INIT;
2128 cache_tree_write(&sb, istate->cache_tree);
2129 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2130 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2131 strbuf_release(&sb);
2132 if (err)
2133 return -1;
2135 if (!strip_extensions && istate->resolve_undo) {
2136 struct strbuf sb = STRBUF_INIT;
2138 resolve_undo_write(&sb, istate->resolve_undo);
2139 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2140 sb.len) < 0
2141 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2142 strbuf_release(&sb);
2143 if (err)
2144 return -1;
2146 if (!strip_extensions && istate->untracked) {
2147 struct strbuf sb = STRBUF_INIT;
2149 write_untracked_extension(&sb, istate->untracked);
2150 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2151 sb.len) < 0 ||
2152 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2153 strbuf_release(&sb);
2154 if (err)
2155 return -1;
2158 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2159 return -1;
2160 istate->timestamp.sec = (unsigned int)st.st_mtime;
2161 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2162 return 0;
2165 void set_alternate_index_output(const char *name)
2167 alternate_index_output = name;
2170 static int commit_locked_index(struct lock_file *lk)
2172 if (alternate_index_output)
2173 return commit_lock_file_to(lk, alternate_index_output);
2174 else
2175 return commit_lock_file(lk);
2178 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2179 unsigned flags)
2181 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2182 if (ret)
2183 return ret;
2184 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2185 (COMMIT_LOCK | CLOSE_LOCK));
2186 if (flags & COMMIT_LOCK)
2187 return commit_locked_index(lock);
2188 else if (flags & CLOSE_LOCK)
2189 return close_lock_file(lock);
2190 else
2191 return ret;
2194 static int write_split_index(struct index_state *istate,
2195 struct lock_file *lock,
2196 unsigned flags)
2198 int ret;
2199 prepare_to_write_split_index(istate);
2200 ret = do_write_locked_index(istate, lock, flags);
2201 finish_writing_split_index(istate);
2202 return ret;
2205 static const char *shared_index_expire = "2.weeks.ago";
2207 static unsigned long get_shared_index_expire_date(void)
2209 static unsigned long shared_index_expire_date;
2210 static int shared_index_expire_date_prepared;
2212 if (!shared_index_expire_date_prepared) {
2213 git_config_get_expiry("splitindex.sharedindexexpire",
2214 &shared_index_expire);
2215 shared_index_expire_date = approxidate(shared_index_expire);
2216 shared_index_expire_date_prepared = 1;
2219 return shared_index_expire_date;
2222 static int should_delete_shared_index(const char *shared_index_path)
2224 struct stat st;
2225 unsigned long expiration;
2227 /* Check timestamp */
2228 expiration = get_shared_index_expire_date();
2229 if (!expiration)
2230 return 0;
2231 if (stat(shared_index_path, &st))
2232 return error_errno(_("could not stat '%s"), shared_index_path);
2233 if (st.st_mtime > expiration)
2234 return 0;
2236 return 1;
2239 static int clean_shared_index_files(const char *current_hex)
2241 struct dirent *de;
2242 DIR *dir = opendir(get_git_dir());
2244 if (!dir)
2245 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2247 while ((de = readdir(dir)) != NULL) {
2248 const char *sha1_hex;
2249 const char *shared_index_path;
2250 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2251 continue;
2252 if (!strcmp(sha1_hex, current_hex))
2253 continue;
2254 shared_index_path = git_path("%s", de->d_name);
2255 if (should_delete_shared_index(shared_index_path) > 0 &&
2256 unlink(shared_index_path))
2257 warning_errno(_("unable to unlink: %s"), shared_index_path);
2259 closedir(dir);
2261 return 0;
2264 static struct tempfile temporary_sharedindex;
2266 static int write_shared_index(struct index_state *istate,
2267 struct lock_file *lock, unsigned flags)
2269 struct split_index *si = istate->split_index;
2270 int fd, ret;
2272 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2273 if (fd < 0) {
2274 hashclr(si->base_sha1);
2275 return do_write_locked_index(istate, lock, flags);
2277 move_cache_to_base_index(istate);
2278 ret = do_write_index(si->base, fd, 1);
2279 if (ret) {
2280 delete_tempfile(&temporary_sharedindex);
2281 return ret;
2283 ret = rename_tempfile(&temporary_sharedindex,
2284 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2285 if (!ret) {
2286 hashcpy(si->base_sha1, si->base->sha1);
2287 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2290 return ret;
2293 static const int default_max_percent_split_change = 20;
2295 static int too_many_not_shared_entries(struct index_state *istate)
2297 int i, not_shared = 0;
2298 int max_split = git_config_get_max_percent_split_change();
2300 switch (max_split) {
2301 case -1:
2302 /* not or badly configured: use the default value */
2303 max_split = default_max_percent_split_change;
2304 break;
2305 case 0:
2306 return 1; /* 0% means always write a new shared index */
2307 case 100:
2308 return 0; /* 100% means never write a new shared index */
2309 default:
2310 break; /* just use the configured value */
2313 /* Count not shared entries */
2314 for (i = 0; i < istate->cache_nr; i++) {
2315 struct cache_entry *ce = istate->cache[i];
2316 if (!ce->index)
2317 not_shared++;
2320 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2323 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2324 unsigned flags)
2326 int new_shared_index, ret;
2327 struct split_index *si = istate->split_index;
2329 if (!si || alternate_index_output ||
2330 (istate->cache_changed & ~EXTMASK)) {
2331 if (si)
2332 hashclr(si->base_sha1);
2333 return do_write_locked_index(istate, lock, flags);
2336 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2337 int v = si->base_sha1[0];
2338 if ((v & 15) < 6)
2339 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2341 if (too_many_not_shared_entries(istate))
2342 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2344 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2346 if (new_shared_index) {
2347 ret = write_shared_index(istate, lock, flags);
2348 if (ret)
2349 return ret;
2352 ret = write_split_index(istate, lock, flags);
2354 /* Freshen the shared index only if the split-index was written */
2355 if (!ret && !new_shared_index)
2356 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2358 return ret;
2362 * Read the index file that is potentially unmerged into given
2363 * index_state, dropping any unmerged entries. Returns true if
2364 * the index is unmerged. Callers who want to refuse to work
2365 * from an unmerged state can call this and check its return value,
2366 * instead of calling read_cache().
2368 int read_index_unmerged(struct index_state *istate)
2370 int i;
2371 int unmerged = 0;
2373 read_index(istate);
2374 for (i = 0; i < istate->cache_nr; i++) {
2375 struct cache_entry *ce = istate->cache[i];
2376 struct cache_entry *new_ce;
2377 int size, len;
2379 if (!ce_stage(ce))
2380 continue;
2381 unmerged = 1;
2382 len = ce_namelen(ce);
2383 size = cache_entry_size(len);
2384 new_ce = xcalloc(1, size);
2385 memcpy(new_ce->name, ce->name, len);
2386 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2387 new_ce->ce_namelen = len;
2388 new_ce->ce_mode = ce->ce_mode;
2389 if (add_index_entry(istate, new_ce, 0))
2390 return error("%s: cannot drop to stage #0",
2391 new_ce->name);
2393 return unmerged;
2397 * Returns 1 if the path is an "other" path with respect to
2398 * the index; that is, the path is not mentioned in the index at all,
2399 * either as a file, a directory with some files in the index,
2400 * or as an unmerged entry.
2402 * We helpfully remove a trailing "/" from directories so that
2403 * the output of read_directory can be used as-is.
2405 int index_name_is_other(const struct index_state *istate, const char *name,
2406 int namelen)
2408 int pos;
2409 if (namelen && name[namelen - 1] == '/')
2410 namelen--;
2411 pos = index_name_pos(istate, name, namelen);
2412 if (0 <= pos)
2413 return 0; /* exact match */
2414 pos = -pos - 1;
2415 if (pos < istate->cache_nr) {
2416 struct cache_entry *ce = istate->cache[pos];
2417 if (ce_namelen(ce) == namelen &&
2418 !memcmp(ce->name, name, namelen))
2419 return 0; /* Yup, this one exists unmerged */
2421 return 1;
2424 void *read_blob_data_from_index(const struct index_state *istate,
2425 const char *path, unsigned long *size)
2427 int pos, len;
2428 unsigned long sz;
2429 enum object_type type;
2430 void *data;
2432 len = strlen(path);
2433 pos = index_name_pos(istate, path, len);
2434 if (pos < 0) {
2436 * We might be in the middle of a merge, in which
2437 * case we would read stage #2 (ours).
2439 int i;
2440 for (i = -pos - 1;
2441 (pos < 0 && i < istate->cache_nr &&
2442 !strcmp(istate->cache[i]->name, path));
2443 i++)
2444 if (ce_stage(istate->cache[i]) == 2)
2445 pos = i;
2447 if (pos < 0)
2448 return NULL;
2449 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2450 if (!data || type != OBJ_BLOB) {
2451 free(data);
2452 return NULL;
2454 if (size)
2455 *size = sz;
2456 return data;
2459 void stat_validity_clear(struct stat_validity *sv)
2461 free(sv->sd);
2462 sv->sd = NULL;
2465 int stat_validity_check(struct stat_validity *sv, const char *path)
2467 struct stat st;
2469 if (stat(path, &st) < 0)
2470 return sv->sd == NULL;
2471 if (!sv->sd)
2472 return 0;
2473 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2476 void stat_validity_update(struct stat_validity *sv, int fd)
2478 struct stat st;
2480 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2481 stat_validity_clear(sv);
2482 else {
2483 if (!sv->sd)
2484 sv->sd = xcalloc(1, sizeof(struct stat_data));
2485 fill_stat_data(sv->sd, &st);